The first objective of the chapter is to determine the porosity of the dolomite gravel within an experimental filter; while the second objective is to determine
Column 8 horizontal slices for flow scan Flow
Resulting flow image a
b
c
___________________________________________________________________________________
113
if biofilm growth within this filter could be imaged and quantified in terms of volume and reduced porosity of the filter. Thus, the water and solid (non-water) fraction of the image needs to be segregated from each other using a binary process whereby regions of MR signal (water) image white (1), whilst regions of no signal (solid) image black (0). As it is uncertain as to whether biofilm would return a signal or not (due to variable water content of the bioform), comparison of the Clean and Bio binary images (i.e. before and after growth) was used to determine if there is a third distinguishable fraction of biofilm. While there are innumerable processes of segmentation available (e.g. Sezgin and Sankur (2004);
Kaestner et al. (2008); Iassonov et al. (2009)), including difficult, time consuming manual processes and some automated algorithms for many similar applications, all are inherently user subjective due to differing pre- and post- processing steps, different acquisition of images and varying programs to process the images. Yet, there is not a standard process for segmentation in image processing. Thus, the protocol utilized and outlined in this section was that of Minto (2013) who used an identical experimental set-up and MR machine to that of the present thesis. Minto (2013) utilized an exhaustive evaluation procedure to determine the best overall segmentation method out of numerous pre- processing, segmentation and post-processing combinations. In short, the key rationale and benefits of Minto’s (2013) process include: qualitative assessment of segmentation quality based on area thresholded, level of noise and separation of particles with further analysis of methods compared against experimentally determined bulk porosity, manual segmentation of slices and manual segmentation of the mesh screen of known value. This evaluation ultimately determined the segmentation process utilized within this chapter to give the best quality results given the experimental setup and MR images acquired.
The segmentation process used to analyze the clean scans of all experiments was done in the image processing software ImageJ with the Bruker plugin installed.
The following is a step-by-step process which first creates the useable portion of the 3D scan resulting in the area used to threshold (segment) the water and non- water fractions which is further used to determine porosity and for comparison purposes.
1. The 2Dseq file contains the high resolution MR scan. Once opened in ImageJ it is displayed as 433 horizontal slices which is then resliced to
Chapter 4 MRI of biofouling and metal transport in a gravel filter ___________________________________________________________________________________
114
show the 600 vertical slices (example of grayscale and color are shown in Figure 4.13a and 4.13b) which allows for visualization of the tapering near the outlet end and the start at the diffuser plate.
Figure 4.13. Example of vertical slices in (a) grayscale (b) colour
2. Before and after scans are matched up as best as possible though are subject to error due to placement within the MR bore between experiments (see Section 4.5.5 summarizing uncertainty). Despite the potential small error, the before-after analysis is essential to review temporal changes which have occurred within the column; hence a substack (Fig. 4.14) is made of the useable portion of the 152 vertical scans (see Section 4.2.6) termed the region of interest (ROI). In all cases 3 slices in the middle of the scan are taken out due to an inherent signal interference from the MR instrument visible in all high resolution images.
Figure 4.14. 3D visualization of useable ROI of experimental gravel filter
___________________________________________________________________________________
115
3. The image was segmented by using a thresholding plugin called ‘Auto Local Threshold’ in which each pixel is thresholded individually according to a radius around it. The specific thresholding option used for this data was the Niblack method with a radius of 20, which follows the algorithm:
pixel = ( pixel > mean + k * standard_deviation - c) ? object : background A radius of 20 was chosen since both water and gravel was needed within a region for the algorithm to work correctly. Since the gravel was ~10mm or ~34 pixels across and thus a 17 pixel diameter, a radius of 20 ensured that both water and gravel fractions would be within a region (and not only one or the other) for the algorithm to segment.
4. Once the image has been thresholded, noise outliers are automatically removed for the dark and bright pixels using nearest neighbour functions, so as to remove any erroneous data that may affect the porosity value.
The edges of the column are also cropped so as to not include any edge effects in the porosity measurements. The resulting image is shown in Figure 4.15 in which the segmentation of the water fraction is seen as white (assigned a value of 255) while the non-water (gravel) fraction is seen as black (assigned a value of 0)
Figure 4.15. Middle slice of 3D scan thresholded in ImageJ.
5. Following the segmentation process, the porosity is determined by measuring the percentage area (%area) in ImageJ. This returns a list of the porosity measurements for each of the 152 slices in which the percentage of white pore spaces to black grains of gravel is calculated.
This fulfils our first aim of determining the porosity of a gravel filter column via MRI.
Chapter 4 MRI of biofouling and metal transport in a gravel filter ___________________________________________________________________________________
116
Subsequently, the thresholded high resolution scans are used to determine differences between the before Clean scans and the after Biofilm scans of the four Bio experiments. In ImageJ a thresholded image is segmented into two values of 0 (black) and 255 (white) as seen in Figure 4.15. So to determine differences between two images the following methodology is employed:
(i) the initial clean stack of slices is divided by 2 such that the water-filled pore areas give a value of 127 (coloured green, Figure 4.16a);
(ii) the final Bio stack is divided by 4 to give water-filled pore- spaces a value of 64 (coloured blue, Figure 4.16b). The uncropped images are used for this step to ensure any biofilm growing at the edges were captured
Figure 4.16. (a) Middle slice of a clean thresholded stack divided by 2 (b) middle slice of a biofilm thresholded stack divided by 4
6. Thus, by adding the two divided images together we are able to determine four regions (Figure 4.17): 0+0=0 indicates gravel in both images (black), 0+64=64 indicates gravel in the clean image but water in the biofilm image (blue), 127+0=127 indicates water in the clean image but solid in the biofilm image (green) and 127+64=191 indicates water in both images (white). Thus, values of 64 (blue) may indicate local movement of gravels away from that location due to settlement/rotation in the column in a manner appropriate to enhanced packing under fluid pore pressures. Similarly, values of 127 (green) may indicate local movement of gravels into that location or potential biofilm growth.
Deciphering the green regions in more detail forms subsequent analysis in this Chapter (Section 4.3) and seeks to fulfil the second aim of
___________________________________________________________________________________
117
determining the viability of using MRI to image biofilm growth within a gravel filter column.
Figure 4.17. Resulting image from adding Clean ÷ 2 with Bio ÷ 4
7. In order to determine the percentage of green versus blue regions for each experiment, a region of interest of the initial 152 slices (with 3 middle interference slices removed) was reduced to 140 slices to eliminate the mesh screen area near the inlet. These 140 slices were then used to obtain the total number of pixels of the stack of images for each colour; green, blue, black and white. Percentages of each specified colour region were then determined from comparing against the total pixels of the circular column area.